India looks towards a DPI approach to open up AI infrastructure

The Telegana Data Exchange (TGDeX) was India’s first digital public infrastructure (DPI) for artificial intelligence (AI) being developed by the state government of Telangana, India. 

The data exchange integrates government, academic and private sector data sets into one platform, with the aim of creating 2,000 AI-ready datasets in five years from 2025 to 2030.  

The convergence of DPI and AI has been underlined in the “Democratising access to AI infrastructure” white paper published on December 29, 2025, by the Office of the Principal Scientific Adviser to the Government of India.  

"This enables multiple stakeholders to collaborate on AI development while maintaining data sovereignty, as it provides secure and privacy-compliant sharing of datasets without requiring movement of raw data,” the whitepaper noted. 

The white paper has been prepared after consultations, policy engagements and expert reviews with key stakeholders in the AI ecosystem, and intended to shape India’s AI policy and governance landscape.  

It highlighted that democratising access to AI infrastructure is “a policy priority for India”. 

“For India, democratising access means treating these building blocks as shared resources so that innovators everywhere can participate in shaping the AI age,” the paper noted, highlighting concerns around the concentration of AI resources among global firms resulting in unequal participation of AI use.  

One of the key enablers put forth by the subcommittee was the need to integrate AI with DPI similar to TGDeX.  

To subscribe to the GovInsider bulletin, click here.

How DPI democratises AI access 

Adopting a DPI approach for AI implies establishing a set of modular public goods that tackle the gaps in the AI ecosystem.  

A key step, according to the paper, is identifying the technical pathways for building an integrated stack that connects foundational AI infrastructure layers, such as computing power and data.

The next step would be to then develop a shared technical architecture to unify these layers to reduce fragmentation and make the use of computing power and datasets more seamless. 

For a start, governments could focus on “lighter weight elements” such as directories, metadata standards, access protocols, or registries.  

“More advanced elements” would range from federated data access systems, consent-based data flows, to coordinated compute-exchange mechanisms.  

“DPI for AI should evolve in a phased and modular manner,” it highlighted. 

Adopting a DPI approach could help to standardise interfaces for interoperability, establish common governance norms which would create transparency, and reduce costs of using AI infrastructure. 

“Its value lies in creating predictable, transparent and interoperable access pathways, particularly for smaller firms, research institutions, and startups that face prohibitive entry barriers,” it said. 

Because of the exponential nature of AI, the paper reinforced the need to roll out a DPI approach for AI in modular phases and position it complementary to other ecosystem interventions, including government investments in AI capacity building and infrastructure expansion.  

The paper also emphasised on the importance of reusable, open layers (or digital public goods) that lower the barriers of entry and enable innovators across the country to build, test, and deploy AI.

These digital public goods that contribute to more accessible AI systems range from open data repositories, subsidised compute clouds, and open-source model hubs. 

You can read the full paper here.